Modeling user's temporal dynamic profile in micro-blogging using clustering method

To help micro-blogging's users find what they want, we need know what user's interests are, but user's interests are not static but change with time, so it is critical to describe what user's current interests are. The micro-blogging messages published by a user contain the user's interest information; intuitively, the message that was published recently by a user should have a bigger impact on the user's current interest than a message that was published long time ago. This paper uses Latent Dirichlet Allocation to extract topics from all messages posted by a user; then, clusters the user's all messages, and represent user's interests with the centers of the clusters. When clustering objects, if apply a time-varying influence function on each object, each object has different influence on clustering process; we call this clustering process as clustering method with weight. The experiment shows that the centers of clusters generated by clustering method with weight can depict user's temporal dynamic profile more accurately than the centers of clusters generated by clustering method without weight.